The Two-Week Assessment That Stopped a Bad Build

Summary

A professional training firm asked us to build a course-recommendation chatbot — immediately, before the busy season. Instead of taking the build, we proposed a two-week assessment of one workflow first. It ended with a written recommendation not to build. They thanked us, fixed the data, and came back six months later. This is the engagement we are proudest of, and it earned us the least.

The Challenge

The Problem

Leadership had promised the board an AI feature for the season. The reality underneath:

• Four disconnected systems: Enrollments, completions, feedback, and billing each lived apart, matched by email address — sometimes.
• No definition of good: Nobody could say what a correct recommendation looked like, so nobody could say whether the chatbot worked.
• A hard deadline: The board announcement set a date before anyone checked feasibility.

Root Cause Analysis

The request was a solution in search of readiness. Building on that foundation would have produced confident, wrong recommendations to paying learners — the kind of failure that costs trust, not just money. The honest work was proving that quickly and cheaply.

Our Approach

Phase 1: Shadow and Profile (Days 1–5)

• Sat with two course coordinators through real recommendation work.
• Profiled all four systems: match rates, freshness, gaps.
• Found that barely half of learner records could be reliably joined.

Phase 2: Score and Recommend (Days 6–10)

• Scored the chatbot against our standard candidate criteria — it failed data readiness outright.
• Wrote the recommendation memo: defer the build, fix identity matching first, revisit in two quarters.
• Presented it to leadership with the numbers, including what the build would have wasted.

Key Achievements

Quantitative Results

• A five-figure build avoided before it spent a dollar of engineering time.
• Two weeks, fixed scope, fixed price — the whole cost of learning the truth early.
• Six months later, the client returned with joined data and a build that succeeded.

Qualitative Benefits

• Leadership walked into the board meeting with facts instead of a delayed project.
• The coordinators got their data fixed — the thing they had actually needed all along.
• The relationship survived a "no," which is how you know it is real.

Lessons Learned

1. The cheapest build is the one you don't start: Two weeks of assessment beat six months of rescue.
2. Bring numbers to a hope: "Barely half the records join" ends debates that opinions cannot.
3. A declined engagement is a deliverable: The memo was the product, and it worked.
4. No is a growth strategy: They came back — with cleaner data and bigger trust.

Conclusion

We tell this story first because it shows how we work when it costs us. Any firm will build what you ask for; few will spend two weeks proving you should not. If that sounds like the partner you want, we should talk.

---

Client: A professional training firm Duration: 2 weeks Team Size: 1 consultant Outcome: An avoided bad build — and a client for years